Introduction

This chapter begins with a quick recap of what natural language processing is and what services it can help provide. Then, it discusses two applications of natural language processing: Parts of Speech Tagging (POS tagging) and Named Entity Recognition. The functioning, necessity, and purposes of both of these algorithms are explained. Additionally, there are exercises and activities that perform POS tagging and named entity recognition and build and develop these algorithms.

Natural language processing consists of aiding machines to understand the natural language of humans in order to communicate with them effectively and automate a large number of tasks. The previous chapter discussed the applications of natural language processing along with examples of real-life use cases where these techniques could simplify the lives of humans. This chapter will specifically look into two of these algorithms and their real-life applications.

Every aspect of natural language processing can...

POS Tagging

Before we dive straight into the algorithm, let's understand what parts of speech are. Parts of speech are something most of us are taught in our early years of learning the English language. They are categories assigned to words based on their syntactic or grammatical functions. These functions are the functional relationships that exist between different words.

Parts of Speech

The English language has nine main parts of speech:

• Nouns: Things or people
• Examples: table, dog, piano, London, towel
• Pronouns: Words that replace nouns
• Examples: I, you, he, she, it
• Verbs: Action words
• Examples: to be, to have, to study, to learn, to play
• Adjectives: Words that describe nouns
• Examples: intelligent, small, silly, intriguing, blue
• Determiners: Words that limit nouns
• Examples: a few, many, some, three

  Note

  For more examples of determiners, visit https://www.ef.com/in/english-resources/english-grammar/determiners/.

• Adverbs: Words that describe...

Applications of Parts of Speech Tagging

Just like text pre-processing techniques help the machine understand natural language better by encouraging it to focus on only the important details, POS tagging helps the machine actually interpret the context of text and thus make sense of it. While text pre-processing is more of a cleaning phase, parts of speech tagging is actually the part where the machine is beginning to output valuable information about corpora on its own.

Understanding what words correspond to which parts of speech can be beneficial in processing natural language in several ways for a machine:

• POS tagging is useful in differentiating between homonyms – words that have the same spelling but mean different things. For example, the word "play" can mean the verb to play, as in engage in an activity, and also the noun, as in a dramatic work to be performed on stage. A POS tagger can help the machine understand what context the word "play" is being...

Chunking

POS taggers work on individual tokens of words. Tagging individual words isn't always the best way to understand corpora, though. For example, the words 'United' and 'Kingdom' don't make a lot of sense when they're separated, but 'United Kingdom' together tells the machine that this is a country, thus providing it with more context and information. This is where the process of chunking comes into the picture.

Chunking is an algorithm that takes words and their POS tags as input. It processes these individual tokens and their tags to see whether they can be combined. The combination of one or more individual tokens is known as a chunk, and the POS tag assigned to such a chunk is known as a chunk tag.

Chunk tags are combinations of basic POS tags. They are easier to define phrases by and are more efficient than simple POS tags. These phrases are chunks. There will be instances where a single word is considered a chunk and assigned a chunk...

Chinking

Chinking is an extension of chunking, as you've probably guessed already from its name. It's not a mandatory step in processing natural language, but it can be beneficial.

Chinking is performed after chunking. Post chunking, you have chunks with their chunk tags, along with individual words with their POS tags. Often, these extra words are unnecessary. They don't contribute to the final result or the entire process of understanding natural language and thus are a nuisance. The process of chinking helps us deal with this issue by extracting the chunks, and their chunk tags form the tagged corpus, thus getting rid of the unnecessary bits. These useful chunks are called chinks once they have been extracted from the tagged corpus.

For example, if you need only the nouns or noun phrases from a corpus to answer questions such as "what is this corpus talking about?", you would apply chinking because it would extract just what you want and present it in front of your...

Named Entity Recognition

This is one of the first steps in the process of information extraction. Information extraction is the task of a machine extracting structured information from unstructured or semi-structured text. This furthers the comprehension of natural language by machines.

After text preprocessing and POS tagging, our corpus becomes semi-structured and machine-readable. Thus, information extraction is performed after we've readied our corpus.

The following diagram is an example of named entity recognition:

Figure 2.12: Example for named entity recognition

Named Entities

Named entities are real-world objects that can be classified into categories, such as people, places, and things. Basically, they are words that can be denoted by a proper name. Named entities can also include quantities, organizations, monetary values, and many more things.

Some examples of named entities and the categories they fall under are as follows:

• Donald Trump, person
• Italy, location...

Summary

Natural language processing enables a machine to understand the language of humans, and just as we learned how to comprehend and process language, machines are taught as well. Two ways of better understanding language that allow machines to contribute to the real world are POS tagging and named entity recognition.

The former is the process of assigning POS tags to individual words so that the machine can learn context, and the latter is recognizing and categorizing named entities to extract valuable information from corpora.

There are distinctions in the way these processes are performed: the algorithms can be supervised or unsupervised, and the approach can be rule-based or stochastic. Either way, the goal is the same, that is, to comprehend and communicate with humans in their natural language.

In the next chapter, we will be discussing neural networks, how they work, and how they can be used for natural language processing.